3 ms·

All these restrictions mean I won't pay for ChatGPT which is a shame because it is powerful. You.com's Chat seems to have less restrictions.

giancarlostoro 4y ago
I think they should just feature flag the uncensored version so you can toggle it, but have to ensure you expunge them of any legal liabilities or whatever. I should be able to ask ChatGPT to make me regexes of known slurs so I can use said regexes for content moderation, but nope, it thinks telling me slurs under any context is the antichrist.
- giraffe_lady 4y ago
  Much more likely their sense of liability extends beyond the strictly legal and they're accounting for the obvious fact that most people requesting that feature will be bad actors. When a headline pops up of "teen bullied into suicide by chatgpt-generated racial tirade" is anyone really going to care that it made their bully check the "we are not legally liable" box before generating it?
  - nivenkos 4y ago
    I think the real risk of bad actors is with spam though - there are already hordes of fake crypto scammers on dating apps, job boards, housing listings, second-hand sales, etc. - now they can use this to generate fake profiles, descriptions and conversations at an even greater scale.
    - giraffe_lady 4y ago
      Sure but that's completely orthogonal from whether it should give you racial slurs lol.
  - giancarlostoro 4y ago
    At that point you might as well do better filtering so that ChatGPT doesn't even actually know the underlying data, ever. Otherwise you'll always have an "AI Jailbreak" until the dawn of time. I think you could paywall it and be surprised by the outcomes. Either way, I was sad that something I helped a friend manually build (regexes of bad words) could never be automated by a simple AI due to people being awful.
  - mikrotikker 4y ago
    Those people won't pay for it. In any case they did the action, they should get the charge. Do the vehicle manufacturers get sued when someone uses them to mow down pedestrians. How about I sue McDonald's after my mother drinks 17 thick shakes back to back and her colon explodes?

giancarlostoro 4y ago

I think they should just feature flag the uncensored version so you can toggle it, but have to ensure you expunge them of any legal liabilities or whatever. I should be able to ask ChatGPT to make me regexes of known slurs so I can use said regexes for content moderation, but nope, it thinks telling me slurs under any context is the antichrist.

giraffe_lady 4y ago
Much more likely their sense of liability extends beyond the strictly legal and they're accounting for the obvious fact that most people requesting that feature will be bad actors. When a headline pops up of "teen bullied into suicide by chatgpt-generated racial tirade" is anyone really going to care that it made their bully check the "we are not legally liable" box before generating it?
- nivenkos 4y ago
  I think the real risk of bad actors is with spam though - there are already hordes of fake crypto scammers on dating apps, job boards, housing listings, second-hand sales, etc. - now they can use this to generate fake profiles, descriptions and conversations at an even greater scale.
  - giraffe_lady 4y ago
    Sure but that's completely orthogonal from whether it should give you racial slurs lol.
- giancarlostoro 4y ago
  At that point you might as well do better filtering so that ChatGPT doesn't even actually know the underlying data, ever. Otherwise you'll always have an "AI Jailbreak" until the dawn of time. I think you could paywall it and be surprised by the outcomes. Either way, I was sad that something I helped a friend manually build (regexes of bad words) could never be automated by a simple AI due to people being awful.
- mikrotikker 4y ago
  Those people won't pay for it. In any case they did the action, they should get the charge. Do the vehicle manufacturers get sued when someone uses them to mow down pedestrians. How about I sue McDonald's after my mother drinks 17 thick shakes back to back and her colon explodes?